fdschmidt93 commited on
Commit
0b238ab
·
1 Parent(s): ef940c3

docs: add README

Browse files
README.md ADDED
@@ -0,0 +1,91 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # MVL-SIB: Massively Multilingual Visual-Language SIB
2
+
3
+ ## Introduction
4
+
5
+ MVL-SIB is a multilingual dataset that provides image-sentence pairs spanning 205 languages and 7 topical categories (`entertainment`, `geography`, `health`, `politics`, `science`, `sports`, `travel`). It was constructed by extending the [SIB-200](https://huggingface.co/datasets/Davlan/sib200) benchmark. For each topic, a set of [10 permissively licensed images](https://huggingface.co/datasets/WueNLP/mvl-sib200/tree/main/data/images/sib200) was manually collected to distinctly represent each category. The dataset creates three instances per original sentence, pairing it with multiple positive and negative image-sentence combinations to challenge both multimodal reasoning and language understanding. MVL-SIB supports detailed evaluations across text-only and cross-modal tasks.
6
+
7
+ ## Usage Example
8
+
9
+ Below is an example of how to load and use the MVL-SIB dataset with the Hugging Face `datasets` library in Python:
10
+
11
+ ```python
12
+ from datasets import load_dataset
13
+
14
+ # Load the MVL-SIB dataset for the 'img2sent' (or 'sent2img') task in English
15
+ dataset = load_dataset("wuenlp/mvl-sib200", name="img2sent.eng_Latn", trust_remote_code=True)
16
+
17
+ print(dataset[0])
18
+ {'images': ['.cache/huggingface/hub/datasets--wuenlp--mvl-sib200/snapshots/96384481f8688607140d69ca45de30cdb18c8596/data/images/sib200/health_5.jpg',
19
+ '.cache/huggingface/hub/datasets--wuenlp--mvl-sib200/snapshots/96384481f8688607140d69ca45de30cdb18c8596/data/images/sib200/health_3.jpg',
20
+ '.cache/huggingface/hub/datasets--wuenlp--mvl-sib200/snapshots/96384481f8688607140d69ca45de30cdb18c8596/data/images/sib200/health_4.jpg',
21
+ '.cache/huggingface/hub/datasets--wuenlp--mvl-sib200/snapshots/96384481f8688607140d69ca45de30cdb18c8596/data/images/sib200/health_6.jpg',
22
+ '.cache/huggingface/hub/datasets--wuenlp--mvl-sib200/snapshots/96384481f8688607140d69ca45de30cdb18c8596/data/images/sib200/health_1.jpg'],
23
+ 'sentences': ['Der „typische” Besuch beinhaltet die Flugreise zum internationalen Flughafen von Orlando, dann die Busfahrt zu einem Disney-Hotel auf dem Gelände, danach einen etwa wochenlangen Aufenthalt dort, ohne das Disney-Gelände zu verlassen, und anschließend die Heimreise.',
24
+ 'Das KI-System wird heute häufig in den Bereichen Wirtschaft, Medizin, Ingenieurwesen und Militär eingesetzt und ist zudem in zahlreiche Softwareanwendungen für Heimcomputer und Videospiele eingebaut worden.',
25
+ 'Am Montag haben die Wisenschaftler der Stanford University School of Medicine die Erfindung eines neuen Diagnosetools bekanntgegeben, mit dem Zellen nach ihrem Typ sortiert werden können: ein winziger, ausdruckbarer Chip, der für jeweils etwa einen US-Cent mit Standard-Tintenstrahldruckern hergestellt werden kann.',
26
+ '1895 unterzeichnete die Regierung der Qing-Dynastie nach der Niederlage im ersten Chinesisch-Japanischen Krieg (1894-1895) den Vertrag von Shimonoseki, in dem sie die Souveränität über Taiwan an Japan abtrat, welches die Insel bis 1945 regierte.'],
27
+ 'categories': ['travel', 'science', 'health', 'politics'],
28
+ 'label': 2,
29
+ 'id': 0,
30
+ 'index_id': 0}
31
+ ```
32
+
33
+
34
+ ## Tasks
35
+
36
+ Large vision-language models must select one of 4 candidate sentences that best matches the topic of the reference images (\`images-to-sentence') or, conversely, choose one of 4 candidate images corresponding to the topic of the reference sentences (\`sentences-to-image'). We present the model with the list of topics that images and sentences may be associated with. Otherwise, it would be unclear along which dimension the model should match images and sentences. The portion of the prompt that introduces the task is provided in English, while the sentences to be topically aligned with images are presented in one of the 205 languages included in MVL-SIB.
37
+
38
+ ### Images-To-Sentence
39
+
40
+ ![Images-To-Sentence](./mvl-sib_images-to-sentence.png)
41
+
42
+ #### Suggested Prompt
43
+
44
+ ```markdown
45
+ Which sentence best matches the topic of the images? The images and the sentences each belong
46
+ to one of the following topics: "entertainment", "geography", "health", "politics", "science and technology", "sports", or "travel". Choose one sentence from A, B, C, or D. Output only a single letter!
47
+
48
+ # Images
49
+ \<IMG\_TOKENS\>
50
+ \<IMG\_TOKENS\>
51
+ \<IMG\_TOKENS\>
52
+ \<IMG\_TOKENS\>
53
+ \<IMG\_TOKENS\>
54
+
55
+ # Sentences
56
+
57
+ A. \`\`\`Maroochydore führte am Ende die Rangfolge an, mit sechs Punkten Vorsprung vor Noosa als Zweitem.\`\`\`
58
+ B. \`\`\`Es wurden keine schwere Verletzungen gemeldet, jedoch mussten mindestens fünf der zur Zeit der Explosion Anwesenden aufgrund von Schocksymptomen behandelt werden.\`\`\`
59
+ C. \`\`\`Finnland ist ein großartiges Reiseziel für Bootstouren. Das „Land der tausend Seen“ hat auch Tausende von Inseln – in den Seen und in den Küstenarchipelen.\`\`\`
60
+ D. \`\`\`Es ist auch nicht erforderlich, dass Sie eine lokale Nummer von der Gemeinde erhalten, in der Sie leben. Sie können eine Internetverbindung über Satellit in der Wildnis v on Chicken in Alaska erhalten und eine Nummer auswählen, die vorgibt, dass Sie im sonnigen Arizona
61
+ sind.\`\`\`
62
+
63
+ Your answer letter:
64
+ ```
65
+
66
+ ### Sentences-To-Image
67
+
68
+ ![Sentences-To-Image](./mvl-sib_sentences-to-image.png)
69
+
70
+ #### Suggested Prompt
71
+
72
+ ```markdown
73
+ Which image best matches the topic of the sentences? The sentences and the images each belong to one of the following topics: "entertainment", "geography", "health", "politics", "science and technology", "sports", or "travel". Choose one image from A, B, C, or D. Output only a single letter!
74
+
75
+ # Sentences
76
+
77
+ - \`\`\`Maroochydore führte am Ende die Rangfolge an, mit sechs Punkten Vorsprung vor Noosa
78
+ als Zweitem.\`\`\`
79
+ - \`\`\`Die Schlagmänner der mittleren Reihe, Sachin Tendulkar und Rahul Dravid, zeigten gute Leistungen und erzielten eine Partnerschaft mit 100 Runs.\`\`\`
80
+ - \`\`\`Da pro Tag nur achtzehn Medaillen zur Verfügung stehen, hat es ein Anzahl an Ländern nicht auf das Podium geschafft.\`\`\`
81
+ - \`\`\`Wintersportarten sind in den nördlichen Regionen am beliebtesten und Italiener nehmen an internationalen Wettkämpfen und olympischen Spielen teil.\`\`\`
82
+ - \`\`\`Nach dem Rennen bleibt Keselowski mit 2.250 Punkten Spitzenreiter in der Fahrerwertung\`\`\`
83
+
84
+ # Images
85
+
86
+ A. \<IMG\_TOKENS\>
87
+ B. \<IMG\_TOKENS\>
88
+ C. \<IMG\_TOKENS\>
89
+ D. \<IMG\_TOKENS\>
90
+ ```
91
+
mvl-sib_images-to-sentence.png ADDED
mvl-sib_sentences-to-image.png ADDED